Resampling Method Research Articles

A Distribution-Preserving Under-Sampling Method for Imbalance Defect Recognition in Castings

Data imbalance is a crucial factor that limits the performance of automatic defect recognition systems in castings. The bias and deterioration of the model are generated by massive normal samples and minor defect samples. Traditional re-sampling methods randomly change the data distribution and ignore the significant intra-class difference among all normal samples. Therefore, this paper proposes a distribution-preserving under-sampling method for imbalance defect-recognition in castings. In detail, our method divides all normal samples into several sub-groups by cluster analysis and reassembles them into some balance datasets, which makes the normal samples in all balance datasets have an identical distribution with the original imbalance dataset. Finally, experiments on our dataset with 3260 images indicate that the proposed method achieves a 0.816 AUC (area under curve) score, which demonstrates significant advantages compared to cost-sensitive learning and re-sampling methods.

Monte-Carlo based random passive energy beamforming for reconfigurable intelligent surface assisted wireless power transfer

Reconfigurable intelligent surface (RIS) employs passive beamforming to control the wireless propagation channel, which benefits the wireless communication capacity and the received energy efficiency of wireless power transfer (WPT) systems. Such beamforming schemes are classified as discrete and non-convex integer programming problems. In this paper, we propose a Monte-Carlo (MC) based random energy passive beamforming of RIS to achieve the maximum received power of electromagnetic (EM) WPT systems. Generally, the Gibbs sampling and re-sampling methods are employed to generate phase shift vector samples. And the sample with the maximum received power is considered the optimal solution. In order to adapt to the application scenarios, we develop two types of passive beamforming algorithms based on such MC sampling methods. The first passive beamforming uses an approximation of the integer programming as the initial sample, which is calculated based on the channel information. And the second one is a purely randomized algorithm with the only total received power feedback. The proposed methods present several advantages for RIS control, e.g., fast convergence, easy implementation, robustness to the channel noise, and limited feedback requirement, and they are applicable even if the channel information is unknown. According to the simulation results, our proposed methods outperform other approximation and genetic algorithms. With our methods, the WPT system even significantly improves the power efficiency in the nonline-of-sight (NLOS) environment.

Accurate and efficient measurement of channelized Hotelling observer-based low-contrast detectability on the ACR CT accreditation phantom.

Current CT quality control (QC) for low-contrast detectability relies on visual inspection and measurement of contrast-to-noise ratio (CNR). However, CNR numbers become unreliable when it comes to nonlinear methods, such as iterative reconstruction (IR) and deep-learning-based techniques. Image quality metrics using channelized Hotelling observer (CHO) have been validated to be well correlated with human observer performance on phantom-based and patient-based tasks, but it has not been widely used in routine CT QC mainly because the CHO calculation typically requires a large number of repeated scans in order to provide accurate and precise estimate of index of detectability (d'). The main goal of this work is to optimize channel filters and other CHO parameters and accurately estimate the low-contrast detectability with minimum number of repeated scans for the widely used American College of Radiology (ACR) CT accreditation phantom so that it can become practically feasible for routine CT QC tests. To provide a converged d' value, an ACR phantom was repeatedly scanned 100 times at three dose levels (24, 12, and 6 mGy). Images were reconstructed with two kernels (FBP Br44 and IR Br44-3). d' as a function of number of repeated scans was determined for different number of background regions of interest (ROIs), different number of low-contrast objects, different number of slices per each object, and different channel filter options. A reference d' was established using the optimized CHO setting, and the bias of d' was quantified using the d' calculated from all 100 repeated scans. The variation of d' at each condition was estimated using a resampling method combining random subsampling among 100 repeated scans and bootstrapping of the ensembles of signal and background ROIs. Optimized parameters in CHO calculation were determined: two background ROIs per object, four objects per low-contrast object size, nine non-overlapping slices per object, and a 4-channel Gabor filter. The bias and uncertainty were estimated at different numbers of repeated scans using these parameters. When only one single scan was used in the CHO calculation, the bias of d' was below 6.2% and the uncertainty 15.6-19.6% for the 6, 5, and 4 mm objects, while with three repeated scans the bias was below 2.0% and uncertainty 8.7-10.9% for the three object sizes. With optimized parameter settings in CHO, efficient and accurate measurement of low-contrast detectability on the commonly used ACR phantom becomes feasible, which could potentially lead to adoption of CHO-based low-contrast evaluation in routine QC tests.

A health state prediction method of traction converter IGBT based on optimized particle filter

Insulated gate bipolar transistor (IGBT), as the most critical and vulnerable power electronic component in the field of rail transportation, its reliability has become a research hotspot in recent years. To improve the safe and reliable operation of rail transportation and to reduce the maintenance cost by rationalizing system maintenance, it is critical to monitor the health of the traction converter IGBTs. This paper first reviews the failure mechanism of the IGBT, then analyzes the health state characterization parameters of the IGBT by the accelerated aging test, selects thermal resistance as the health state characterization parameter of the IGBT, and establishes the degradation model of the IGBT. On this basis, a health state prediction method of IGBT based on the genetic algorithm unscented particle filter (GA-UPF) is proposed. This method uses an unscented Kalman filter to generate important proposal density distribution for each particle, estimates, and updates the parameters of the degradation model, alleviates the particle degradation problem, and uses a genetic algorithm to solve the particle impoverishment problem introduced by the traditional resampling method, which improves the particle diversity and prediction accuracy. Mahalanobis distance is used to monitor the abnormal state of the thermal resistance. After the abnormal occurs, the value of the thermal resistance is predicted according to the degradation model and updated parameters. The degradation trend of the thermal resistance is predicted to reflect the health state of the IGBT. Experiments on IGBT degradation data show that the proposed method can provide higher health state prediction accuracy than the traditional particle filter method.

Parkinson's disease resting tremor severity classification using machine learning with resampling techniques.

In resting tremor, the body part is in complete repose and often dampens or subsides entirely with action. The most frequent cause of resting tremors is known as idiopathic Parkinson's disease (PD). For examination, neurologists of patients with PD include tests such as finger-to-nose tests, walking back and forth in the corridor, and the pull test. This evaluation is focused on Unified Parkinson's disease rating scale (UPDRS), which is subjective as well as based on some daily life motor activities for a limited time frame. In this study, severity analysis is performed on an imbalanced dataset of patients with PD. This is the reason why the classification of various data containing imbalanced class distribution has endured a notable drawback of the performance achievable by various standard classification learning algorithms. In this work, we used resampling techniques including under-sampling, over-sampling, and a hybrid combination. Resampling techniques are incorporated with renowned classifiers, such as XGBoost, decision tree, and K-nearest neighbors. From the results, it is concluded that the Over-sampling method performed much better than under-sampling and hybrid sampling techniques. Among the over-sampling techniques, random sampling has obtained 99% accuracy using XGBoost classifier and 98% accuracy using the decision tree. Besides, it is observed that different resampling methods performed differently with various classifiers.

Quantile regression for competing risks data from stratified case-cohort studies: an induced-smoothing approach

ABSTRACT A case-cohort design offers an economical way of investigating an association between exposure variables and risks of disease outcomes compared to a large-scale full cohort study. A stratified sampling in such designs based on the information available for the entire cohort is often considered for improving efficiencies of estimators. In this paper, we consider fitting censored quantile regression models for competing risks data arising from stratified case-cohort designs. We model quantiles for cumulative incidence functions that provide desirable interpretation for competing risks data and flexible ways of assessing covariates effects. For estimation of regression parameters, we consider weighted estimating equations with two-types of weights: the inverses of censoring probabilities and sampling probabilities to account for censoring in competing risks data and biased features in stratified case-cohort samplings, respectively. An induced smoothing approach is applied to obtain computationally more reliable estimates. The resulting estimating functions are smooth in regression parameters, so standard numerical algorithms for point estimation can be readily applied and variances can be estimated via a closed-form expression or computationally efficient resampling method. An iterative algorithm is proposed to simultaneously estimate regression parameters and their variances. Asymptotic properties of the proposed estimators are established. The finite sample properties of the proposed estimators are investigated through extensive simulation studies. They perform reasonably well under practical settings considered. The proposed methods are illustrated with Hodgkin's disease data.

Open world long-tailed data classification through active distribution optimization

Real-world data exhibits a long-tailed label distribution, which leads to classification bias. Popular re-sampling or re-weighting methods usually require known category information. However, learning from long-tailed data with open categories is a challenging issue. In this paper, we propose an active distribution optimization algorithm (DALC) to handle the interesting issue. Through clustering, querying and classification iterations, we explore new categories and balance label distribution. For clustering, we present an exploration technique that adaptively obtains optimal data distribution with minimal total distance/cost. For each query, we design a critical instance selection strategy with the cluster information. For classification, we establish an ensemble model to continuously balance the label distribution. We conducted experiments on synthetic, benchmark and domain datasets. The results of the significance test verified the effectiveness of DALC and its superiority over state-of-the-art long-tailed data classification and open set classification algorithms.

Spatial distribution modeling of customer complaints using machine learning for indoor water leakage management

Customer complaints reflect the needs of citizens and provide valuable information for the efficient management of urban problems. Indoor water leakage management is required to achieve a sustainable water infrastructure and urban development. In this study, a machine learning (ML)-based modeling framework was developed for predicting the spatial distribution of customer complaints about indoor water leakage in the downtown area of Daegu Metropolitan City, South Korea. Two ML algorithms (XGBoost and LightGBM) were used with six resampling methods (e.g., undersampling, oversampling, and hybrid sampling) to compare the prediction performances. The combination of LightGBM and hybrid sampling showed the highest prediction performance. Post hoc analysis using Shapley Additive Explanations indicated that, among the various input features, the land cover type, building and water infrastructure characteristics were of primary importance. High-resolution gridded mapping clearly revealed the spatial pattern of complaint probabilities. These results provide a decision support tool for indoor water leakage management. The proposed modeling framework encompasses data preprocessing and integration, prediction, interpretation, and spatial mapping, and it is applicable to a wide variety of urban problems that require in-depth analysis of their spatial characteristics.

Efficient Sensor Calibration Via Machine Learning-Based Resampling Methods

It remains a challenge to accurately and precisely calibrate a sensor in an experimentally-efficient manner. The calibration task at hand is to quantify the functional relationship between the sensor’s response (denoted as y) and the target analyte concentration (denoted as x) based on laboratory experiments. In practical use, the sensor’s mathematical calibration model needs to be coupled with the device to make inference regarding the concentration of the analyte in an environment based on the observed sensor response. It is critical to provide a single-value estimate for the analyte concentration, as well as to quantify the uncertainty of the target estimate (i.e., to provide the possible range of the underlying concentration level).However, classic statistical methods are not able to provide valid inference on the target quantity of interest due to the following difficulties in sensor calibration. First, not only the typical forward modeling but also the inverse estimation is involved. The former refers to the generation of a calibration model functionally relating the analyte concentration to the sensor response (x→y) based on experimental data; The latter concerns the use of the calibration model in practical setting to obtain from the observed response the estimation of the unknown concentration (y→x). Second, the variability of the sensor responses tends to be heterogeneous as opposed to homogeneous, which is an assumption that classic statistical methods rely on.In light of the above, a computational resampling method was adapted in this work to address the machine learning (ML)-based inference issues (i.e., the uncertainty quantification of the estimated analyte concentrations). Built on the inference capability, a two-stage experimental design procedure was developed to guide the efficient sampling of calibration data via laboratory experiments.The ML-based inference and experimental design methods are integrated in the chemometric calibration procedure, which was applied on a simulated biosensor to demonstrate its efficiency over traditional methods. The sensor simulator is developed from the real experimental data for a paper-based lateral flow strip (PLFS) designed to measure in blood the diagnosis biomarkers for traumatic brain injury. PLFS is finding increasing applications in detection of various diseases in point-of-care testing settings. Figure 1

Quicksort leave-pair-out cross-validation for ROC curve analysis

Receiver Operating Characteristic (ROC) curve analysis and area under the ROC curve (AUC) are commonly used performance measures in diagnostic systems. In this work, we assume a setting, where a classifier is inferred from multivariate data to predict the diagnostic outcome for new cases. Cross-validation is a resampling method for estimating the prediction performance of a classifier on data not used for inferring it. Tournament leave-pair-out (TLPO) cross-validation has been shown to be better than other resampling methods at producing a ranking of data that can be used for estimating the ROC curves and areas under them. However, the time complexity of TLPOCV, Oleft( n^2right), means that it is impractical in many applications. In this article, a method called quicksort leave-pair-out cross-validation (QLPOCV) is presented in order to decrease the time complexity of obtaining a reliable ranking of data to Oleft( nlog nright). The proposed method is compared with existing ones in an experimental study, demonstrating that in terms of ROC curves and AUC values QLPOCV produces as accurate performance estimation as TLPOCV, outperforming both k-fold and leave-one-out cross-validation.

Uncertainty-Constrained Belief Propagation for Cooperative Target Tracking

Cooperative localization is essential for many Internet of Things (IoT)-related applications in harsh environments. Generally, the inertial navigation system is self-contained and adopted as the basis of a cooperative tracking system, but it still faces the problem of accumulated errors and cannot provide long-term, high-precision positioning. The particle filter (PF) is widely used to fuse multiple information to inhibit accumulative errors. However, particle degradation and impoverishment remain unsolved. This article proposed an IMU/time-of-arrival (TOA) fusion-based tracking method, namely, uncertainty-constrained belief propagation (UCBP). We address particle degradation and impoverishment by introducing uncertainty-constrained optimization into belief propagation (BP). An uncertainty-constrained resampling (UCR) method is applied to quantify the uncertainty in cooperative systems. Hierarchical resampling is realized to solve the particle impoverishment issue. Meanwhile, particle degradation is resolved through constrained resampling while ensuring the diversity of particles. Furthermore, we illustrated the factor graph (FG) structure of UCBP to mitigate the accumulation of errors through message fusion over the graph. Compared with the state-of-the-art methods, our proposed UCBP algorithm has better precision and robustness without introducing much time overhead.

Effective hybrid feature selection using different bootstrap enhances cancers classification performance

BackgroundMachine learning can be used to predict the different onset of human cancers. Highly dimensional data have enormous, complicated problems. One of these is an excessive number of genes plus over-fitting, fitting time, and classification accuracy. Recursive Feature Elimination (RFE) is a wrapper method for selecting the best subset of features that cause the best accuracy. Despite the high performance of RFE, time computation and over-fitting are two disadvantages of this algorithm. Random forest for selection (RFS) proves its effectiveness in selecting the effective features and improving the over-fitting problem.MethodThis paper proposed a method, namely, positions first bootstrap step (PFBS) random forest selection recursive feature elimination (RFS-RFE) and its abbreviation is PFBS- RFS-RFE to enhance cancer classification performance. It used a bootstrap with many positions included in the outer first bootstrap step (OFBS), inner first bootstrap step (IFBS), and outer/ inner first bootstrap step (O/IFBS). In the first position, OFBS is applied as a resampling method (bootstrap) with replacement before selection step. The RFS is applied with bootstrap = false i.e., the whole datasets are used to build each tree. The importance features are hybrid with RFE to select the most relevant subset of features. In the second position, IFBS is applied as a resampling method (bootstrap) with replacement during applied RFS. The importance features are hybrid with RFE. In the third position, O/IFBS is applied as a hybrid of first and second positions. RFE used logistic regression (LR) as an estimator. The proposed methods are incorporated with four classifiers to solve the feature selection problems and modify the performance of RFE, in which five datasets with different size are used to assess the performance of the PFBS-RFS-RFE.ResultsThe results showed that the O/IFBS-RFS-RFE achieved the best performance compared with previous work and enhanced the accuracy, variance and ROC area for RNA gene and dermatology erythemato-squamous diseases datasets to become 99.994%, 0.0000004, 1.000 and 100.000%, 0.0 and 1.000, respectively.ConclusionHigh dimensional datasets and RFE algorithm face many troubles in cancers classification performance. PFBS-RFS-RFE is proposed to fix these troubles with different positions. The importance features which extracted from RFS are used with RFE to obtain the effective features.

Efficient Sensor Calibration Via Machine Learning-Based Resampling Methods

A sensor’s performance relies on the calibration model functionally relating the sensor’s response (denoted as y) to the target analyte concentration (denoted as x). However, classic statistical methods are not able to adequately quantify the uncertainty of the x concentration estimates. In this work, a machine learning (ML)-based computational resampling method was adapted to address the uncertainty quantification issues, based on which a two-stage procedure was developed for efficient sampling and modeling of the calibration data obtained via laboratory experiments. The outputs of the procedure allow for not only a single-value estimate, but also a high-likelihood range for the underlying analyte concentration in an unknown sample being tested. The calibration procedure was applied on a simulated biosensor, and evaluated through large amount of validation data. The sensor simulator was developed from a paper-based lateral flow strip (PLFS) designed for traumatic brain injury (TBI) diagnosis.

Development of a nomogram for predicting pelvic lymph node metastasis in cervical squamous cell carcinoma.

To develop and validate a nomogram for predicting pelvic lymph node metastasis (LNM) in cervical squamous cell carcinoma (SCC). This was a retrospective study that included 715 patients with cervical SCC who underwent radical hysterectomy and bilateral pelvic lymphadenectomy between 2009 and 2018. Logistic regression analysis was used to identify independent risk factors for pelvic LNM. Based on these risk factors, a nomogram predicting LNM risk was constructed and internally validated using the bootstrapping resampling method. The rate of LNM in FIGO (the International Federation of Gynecology & Obstetrics) Stage IA2-IIA2 cervical SCC was 24.2%. In multivariate analysis, FIGO Stage II, moderately differentiated or poorly differentiated histology, abnormally elevated serum SCC-antigen, and triglyceride were identified as independent risk factors for LNM. Tumor size greater than 2 cm and parametrial involvement had borderline significance. Ultimately, the nomogram contained the six variables mentioned above, showing positive calibration and positive discrimination. The area under the receiver operating characteristic curvewas 0.827 and the bootstrap-validated C-index was 0.827. The Youden index of this paper was 0.540. We developed and validated a nomogram to predict pelvic LNM in SCC based on clinical data, which can help physicians develop an optimal treatment strategy.

Research on Resampling and Clustering Method of Aircraft Flight Trajectory

Research on Resampling and Clustering Method of Aircraft Flight Trajectory

Fraud Detection Using Large-scale Imbalance Dataset

In the context of machine learning, an imbalanced classification problem states to a dataset in which the classes are not evenly distributed. This problem commonly occurs when attempting to classify data in which the distribution of labels or classes is not uniform. Using resampling methods to accumulate samples or entries from the minority class or to drop those from the majority class can be considered the best solution to this problem. The focus of this study is to propose a framework pattern to handle any imbalance dataset for fraud detection. For this purpose, Undersampling (Random and NearMiss) and oversampling (Random, SMOTE, BorderLine SMOTE) were used as resampling techniques for the concentration of our experiments for balancing an evaluated dataset. For the first time, a large-scale unbalanced dataset collected from the Kaggle website was used to test both methods for detecting fraud in the Tunisian company for electricity and gas consumption. It was also evaluated with four machine learning classifiers: Logistic Regression (LR), Naïve Bayes (NB), Random Forest, and XGBoost. Standard evaluation metrics like precision, recall, F1-score, and accuracy have been used to assess the findings. The experimental results clearly revealed that the RF model provided the best performance and outperformed all other matched classifiers with attained a classification accuracy of 89% using NearMiss undersampling and 99% using Random oversampling.

Effects of classification, feature selection, and resampling methods on bankruptcy prediction of small and medium‐sized enterprises

SummarySmall and medium‐sized enterprises are the pillars of an economy, and their poor performance has a negative impact on living standards of population and country development. This study analyzes real‐life data of 89,851 small and medium‐sized enterprises, out of which 295 have declared bankruptcy. The analysis is performed via 27 financial ratios. The study framework combines seven classifications and three resampling and seven feature selection methods. Out of all classification methods applied, CatBoost has achieved the best results for all combinations of resampling and feature selection methods. CatBoost surpassed the results of other classification methods for the area under curve parameter, achieving a value of 99.95%. The application of resampling methods on different classification models has not identified a statistically significant level of improvement in any of the resampling methods. This finding has also been observed for feature selection methods. Based on these findings, we assume that individual resampling and feature selection methods do not improve model performance compared with the original imbalanced sample's results. Our results suggest that, even though the data sample may be significantly imbalanced with a minority of bankrupt companies, most classification algorithms can handle this imbalance and achieve interesting results. Moreover, our findings provide broad practical application for all stakeholders who could need to detect bankrupting companies.

A New Bayesian Person‐Fit Analysis Method Using Pivotal Discrepancy Measures

AbstractA common method of checking person‐fit in Bayesian item response theory (IRT) is the posterior‐predictive (PP) method. In recent years, more powerful approaches have been proposed that are based on resampling methods using the popular statistic. There has also been proposed a new Bayesian model checking method based on pivotal discrepancy measures (PDMs). A PDM T is a discrepancy measure that is a pivotal quantity with a known reference distribution. A posterior sample of T can be generated using standard Markov chain Monte Carlo output, and a p‐value is obtained from probability bounds computed on order statistics of the sample. In this paper, we propose a general procedure to apply this PDM method to person‐fit checking in IRT models. We illustrate this using the and measures. Simulation studies are done comparing these with the PP method and one of the more recent resampling methods. The results show that the PDM method is more powerful than the PP method. Under certain conditions, it is more powerful than the resampling method, while in others, it is less. The PDM method is also applied to a real data set.

Developing machine learning-based models to predict intrauterine insemination (IUI) success by address modeling challenges in imbalanced data and providing modification solutions for them

BackgroundThis study sought to provide machine learning-based classification models to predict the success of intrauterine insemination (IUI) therapy. Additionally, we sought to illustrate the effect of models fitting with balanced data vs original data with imbalanced data labels using two different types of resampling methods. Finally, we fit models with all features against optimized feature sets using various feature selection techniques.MethodsThe data for the cross-sectional study were collected from 546 infertile couples with IUI at the Fatemehzahra Infertility Research Center, Babol, North of Iran. Logistic regression (LR), support vector classification, random forest, Extreme Gradient Boosting (XGBoost) and, Stacking generalization (Stack) as the machine learning classifiers were used to predict IUI success by Python v3.7. We employed the Smote-Tomek (Stomek) and Smote-ENN (SENN) resampling methods to address the imbalance problem in the original dataset. Furthermore, to increase the performance of the models, mutual information classification (MIC-FS), genetic algorithm (GA-FS), and random forest (RF-FS) were used to select the ideal feature sets for model development.ResultsIn this study, 28% of patients undergoing IUI treatment obtained a successful pregnancy. Also, the average age of women and men was 24.98 and 29.85 years, respectively. The calibration plot in this study for IUI success prediction by machine learning models showed that between feature selection methods, the RF-FS, and among the datasets used to fit the models, the balanced dataset with the Stomek method had well-calibrating predictions than other methods. Finally, the brier scores for the LR, SVC, RF, XGBoost, and Stack models that were fitted utilizing the Stomek dataset and the chosen feature set using the Random Forest technique obtained equal to 0.202, 0.183, 0.158, 0.129, and 0.134, respectively. It showed duration of infertility, male and female age, sperm concentration, and sperm motility grading score as the most predictable factors in IUI success.ConclusionThe results of this study with the XGBoost prediction model can be used to foretell the individual success of IUI for each couple before initiating therapy.

Development of a Machine Learning-Based Screening Method for Thyroid Nodules Classification by Solving the Imbalance Challenge in Thyroid Nodules Data.

This study aims to show the impact of imbalanced data and the typical evaluation methods in developing and misleading assessments of machine learning-based models for preoperative thyroid nodules screening. A retrospective study. The ultrasonography features for 431 thyroid nodules cases were extracted from medical records of 313 patients in Babol, Iran. Since thyroid nodules are commonly benign, the relevant data are usually unbalanced in classes. It can lead to the bias of learning models toward the majority class. To solve it, a hybrid resampling method called the Smote-was used to creating balance data. Following that, the support vector classification (SVC) algorithm was trained by balance and unbalanced datasets as Models 2 and 3, respectively, in Python language programming. Their performance was then compared with the logistic regression model as Model 1 that fitted traditionally. The prevalence of malignant nodules was obtained at 14% (n = 61). In addition, 87% of the patients in this study were women. However, there was no difference in the prevalence of malignancy for gender. Furthermore, the accuracy, area under the curve, and geometric mean values were estimated at 92.1%, 93.2%, and 76.8% for Model 1, 91.3%, 93%, and 77.6% for Model 2, and finally, 91%, 92.6% and 84.2% for Model 3, respectively. Similarly, the results identified Micro calcification, Taller than wide shape, as well as lack of ISO and hyperechogenicity features as the most effective malignant variables. Paying attention to data challenges, such as data imbalances, and using proper criteria measures can improve the performance of machine learning models for preoperative thyroid nodules screening.